Why voter surveys don't agree

The latest Pew poll provides stark evidence that the electorate may be more volatile than previously thought. The survey of 1,972 registered voters was taken in two waves. The first group, polled Sept. 8-10, gave Bush a whopping 54-38 percent lead among likely voters. But by the second group, surveyed Sept. 11-14, that lead had vanished, and the race among likely voters showed Bush up by just 1 point, 47 to 46 percent.

Pew describes crosscurrents in opinion that could push swing voters first one way, then another. The pounding that Kerry took during August and at the GOP convention in New York clearly hurt the Democrat, while Bush remains vulnerable on Iraq and the economy, the Pew report said.

But even with a poll whose top line shows a statistical tie, it's possible to give one side an edge, says Raghavan Mayur, president of the TIPP poll, which conducts surveys for this newspaper. His latest poll, released Sept. 14, showed Bush and Kerry tied at 46 percent each. But Mr. Mayur still sees Bush as slightly ahead, based on the intensity of his voters.

"Whenever there is a tie, what I do is look at the next layer of statistics to see how the wind is blowing," he says.

All polls vary in their methodology, but perhaps none is more subjective than weighting for party identification - a practice some follow and others don't. Over the past three decades, the Democratic lead in voters' self-identification with that party has declined compared with Republicans; some pollsters factor in evolving voter ID when they weight their samples while others don't.

John Zogby, whose eve-of-election polling accurately predicted the final outcome in both the 2000 and 1996 presidential races, does weight his samples according to his best sense of how voters identify themselves. "We all use artistry; the difference is I admit it," says Mr. Zogby. "I'm a historian and when you're dealing with people, it's not all statistics."

Another factor that has made polling increasingly difficult is the public's declining willingness to give pollsters the time of day. The new do-not-call registry does not apply to pollsters, but it has emboldened the public to ignore, or reject with expletives, their requests for interviews. Proliferation of caller ID also hurts pollsters.

The fact that cellphones are off limits to pollsters also hinders outreach into certain demographics, such as young people, who in growing numbers have no land lines.

But young people - as well as minorities - have always been more difficult for pollsters to reach than other groups.

Below are results of three recent nationwide polls of likely voters, showing only responses for George Bush or John Kerry.

The Christian Science Monitor/TIPP (Sept. 7-12)
Kerry: 46%
Bush: 46%

Harris (Sept. 9-13)
Kerry: 48%
Bush: 47%

Gallup (Sept. 13-15)
Kerry: 42%
Bush: 55%

Page: 1 | 2